Release 2.19.2 — `versionName` 2.19.2 / `versionCode` 21902. Merging this cuts the release.
A meeting you declined is struck through everywhere it appears and no longer schedules a reminder (#180). Tapping an event in month view opens the event instead of its day, in every month style (#187). An edit shows immediately when the same occurrence is re-opened (#196).
The month widget's arrows died after a couple of taps: the grid serialised ~740 views and one update ran to 450–540 KB, which overran the launcher's async binder buffer and made Android drop the whole widget host — killing updates for every widget on the home screen. The grid is down to 192 views, and `SizeMode.Exact` is kept so a resized widget reflows instead of clipping (#214). Widget taps now redraw the tapped widget by its own id rather than `updateAll`, whose provider lookup is empty in a cold-woken process (#18).
Jump-to-today in seamless weeks lands on the current week row (#191), and the day and week timelines got an end inset (#192).
Two follow-ups beyond the issues: the detail re-read keeps the loaded content on screen instead of blinking back to the skeleton, and a month chip tap with no recorded touch down (TalkBack, D-pad) no longer resolves to the previous tap's chip.
Release notes: `CHANGELOG.md` 2.19.2 section and `fastlane/metadata/android/en-US/changelogs/21902.txt` (471 chars).
Closes#18Closes#180Closes#187Closes#191Closes#192Closes#196Closes#214
Co-authored-by: Jean-Luc Makiola <business@jeanlucmakiola.de>
Reviewed-on: https://codeberg.org/jlmakiola/calendula/pulls/216
A multi-day event has a dot on every day it covers but only one bar, so the
dots away from its start had no counterpart. Untagged they stood still and
faded; tagged without a partner they flew in from the top of the grid.
Neither is an animation — the last commit swapped one for the other.
The expanded grid now places an invisible slice of the bar in each further
column it spans, keyed to that column's day. Every dot has a real place to
come out of and go back into, at its own column, so it drops out of the bar
above it instead of appearing from nowhere.
That makes the seating's anchored flag pointless again — every dot has a
partner now — so it and its tests come back out.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Dots on the middle of a multi-day run flew in from the top of the grid
instead of fading where they stand.
They were meant to be left unmatched — a bar is drawn once per row, from
the column it starts in, so only that day's dot has a counterpart. But
"unmatched" was implemented as tagged-and-unmatched, which is a different
thing: a shared element entering with no partner has no bounds to start
from, so it animates in from the layout origin. The top.
Seating now says which day a bar is actually drawn from, and only that dot
carries a tag. The rest are plain content and fade in place, which is what
the comment claimed they did all along.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The dots gathered a day's *distinct colours*, which meant dot i stood for
no particular event and two events sharing a calendar collapsed into one —
undercounting the day.
Seat them off the same lane assignment the paged grid draws from, so a dot
and the bar in that lane are the same event. That fixes the count, and it
is what will let one morph into the other when the grid expands.
The cap is now MAX_EVENT_ROWS rather than a constant of its own: the two
have to agree or a dot would have no bar to become.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Opening Continuous stayed slow, and the slowness tracked the number of enabled
calendars rather than the window size — which pointed past the provider query to
the layout behind it.
layoutCalendarWeek opens by filtering the instances it is handed down to the
ones touching its seven days. For a single month's grid that is nothing; for the
scrolling styles it was a scan of every instance in an eleven-month window,
repeated for each of a hundred-odd rows, with a time-zone conversion per check.
The work grew with events × rows, so switching every calendar on multiplied it.
- DayIndex buckets the window's events by the dates they cover, once, and each
row takes the handful on its own seven days. Membership is decided by
coversDay itself rather than re-derived from the timestamps — the all-day and
timed cases have enough edge cases between them (UTC anchoring, exclusive
ends, zero-length events at midnight) that a second implementation would
drift. DayIndexTest pins row-for-row equality with the old full scan, in a
zone east of UTC.
- Only the style on screen is laid out. Adding Dense had quietly doubled the
work, since both layouts were built from every load and one was always thrown
away.
- The paged flow no longer queries under a scrolling style, mirroring the gate
the scrolling flow already had.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Opening Continuous made the first frame wait for nine months of recurrence
expansion when only one was about to be looked at — a provider Instances query
expands across its whole range, so the window's generosity was paid for up
front, every time.
The window now starts at one month either side of the visible one — about what
the paged style costs — and each completed load reaches two months further in
both directions until it spans eleven. The months a scroll can reach arrive
while the first one is already on screen, and because a widening is triggered by
the previous load landing rather than by a timer, the ladder can never outrun
the provider.
- The paged and split styles no longer run this query at all. The screen
collects the flow whatever the style is set to, so until now every Month view
opening paid for a window it would never draw.
- The reload trigger is derived from the current pad instead of being fixed, and
is held strictly inside it: a trigger at or beyond the pad would re-fire the
moment its own reload landed.
- The scrolling styles get their own skeleton — the layout they are about to
become, at the same measurements, so arriving months replace it in place. It
and the per-month placeholders now breathe, so a slow load reads as work
rather than as an empty grid. Held still under reduced motion.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Continuous's month blocks make the boundaries legible, but the seams are exactly
what someone who wants a dense calendar doesn't want. Both are now offered:
- **Dense** is the old flowing layout, kept rather than replaced — one
uninterrupted stream of weeks with months running into each other, the 1st
naming its month, and the top bar keeping the full month title (there is no
sticky header to defer to).
- Continuous gains a **rule under its month label**, closing the header off
against the grid. On trial: it comes out again if it doesn't earn its place.
One provider query serves both. `ContinuousMonthUiState.Success` now carries the
same loaded window laid out twice — `monthsByIndex` clipped into blocks,
`weeksByIndex` left whole — rather than standing up a second flow and querying
the same range again. Both styles report *months* as they scroll, so the window
hysteresis is shared; `weekWindowFor` maps a month window onto the Dense rows it
covers, widened a week each side so boundary rows don't flicker.
The list state is keyed on style *and* week start: Dense indexes by week, which
the week-start preference moves, while Continuous indexes by month, which it
doesn't — carrying an offset between the two would land somewhere arbitrary.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The continuous grid failed outright with "Could not read the calendar" the
moment it opened.
Before the first layout pass LazyListState reports no visible items, and the
screen defaulted that to item 0 — claiming January 1900 was on screen. The load
window widened around it, its pad ran past the epoch to index -4, and
yearMonthForIndex indexed Month.entries out of bounds; the flow's catch turned
the crash into the generic provider failure. Week indexing never showed this:
index 0 was a real week, so a bogus report only cost one wasted query.
- The visible-range report is skipped entirely while the list has no items,
rather than standing in a default for them.
- clampMonthWindow holds any window inside 1900–2100, so a pad at either end
can't produce an index with no month behind it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The continuous style streamed weeks with no boundaries at all, which left it
hard to tell where one month ended: the only marker was a "Jul 1" label on the
1st, and every boundary week mixed two months' days into one row.
It is now a vertical stack of self-contained month blocks. Each block shows only
its own days — the boundary week keeps its seven columns so nothing shifts
sideways, but the neighbour month's cells are blank rather than filled with
duplicates of days shown again a block later — under a sticky month header with
whitespace either side. Scrolling stays continuous; only the reading changes.
- The coordinate space moves from absolute week index to absolute month index
(two LazyColumn items per month: header, then block). Unlike week indices, it
doesn't depend on the week-start preference, so changing that reflows the rows
inside a block without moving the block or losing the scroll position.
- The sliding data window now loads months rather than weeks, widened to whole
grid weeks at both ends so a bar reaching into a block from a clipped-off day
still renders.
- `clipWeekToMonth` is the pure seam: it drops the neighbour month's pills and
counts and cuts spanning bars back to the month's own columns, keeping a flat
cap on the cut side so a bar reads as continuing past the block.
- The top bar carries the year in this style — the block's own header names the
month, so repeating it two lines up was pure duplication.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds MonthViewStyle.Split: the month compressed to day numbers and event dots
over a list of the selected day's events.
- The app's first selected-day concept. Every other view drills straight into a
date; here a tap selects, because the pane below is already the answer to
"what's on this day" and opening a whole screen would defeat the layout. The
full Day view stays one tap away on the pane's date header, matching the Week
and Agenda headers (#37).
- The pane reuses the agenda's row vocabulary (extracted in the groundwork
commit) rather than growing a parallel set, so the two surfaces read as one
app. It needs no extra provider query: instancesByDay is already covered by
the month grid range.
- Selection follows the month — today when the new month holds it, else the 1st
— so the pane never lists a day the grid isn't showing. Tapping a leading or
trailing day follows it to its own month.
- Selection and today are different signals (tinted, outlined cell vs. the
filled circle the other views use for today), so both read when they land on
the same day.
- The swipe stays on the grid alone; the pane scrolls and is full of tappable
rows.
MonthScreen gains onEventClick, wired in CalendarHost alongside the other three
views, and MonthUiState.Success now carries its zone for the same reason the
agenda's does.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds MonthViewStyle.Continuous: one uninterrupted vertical stream of weeks
instead of month pages.
The layout unit is the ISO week, not the month. A vertical stack of month grids
would still repeat a boundary week at the end of one month and the start of the
next, which is precisely the duplication the issue asks to be rid of. Streaming
weeks means every date appears exactly once.
- Weeks are addressed by an absolute index (epoch 1900, so indices stay
non-negative and map 1:1 onto LazyColumn item indices), giving the list one
stable, gap-free coordinate space to scroll and key by.
- The view model loads a sliding window of weeks around the visible range.
nextLoadWindow() holds the hysteresis: the window only moves once the visible
range comes within four weeks of a loaded edge, so scrolling re-queries the
provider occasionally rather than every frame. Unloaded rows render a
same-height skeleton, so nothing jumps when the window catches up.
- No dimmed "other month" days — every day in the stream belongs to a month
equally. Instead the 1st names its month, which is the only marker needed to
tell one month from the next, and the top bar title tracks the month the
viewport mostly sits in.
- The horizontal swipe detector and the paged AnimatedContent are both off in
this style; vertical scrolling owns the gesture. Today and drawer jump-to-date
animate the list instead of swapping months. The view opens positioned on the
current month.
MonthWeekRow now takes inMonth as a predicate rather than a YearMonth, which is
what lets the same row serve a stream that has no enclosing month.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Introduces MonthViewStyle{Paged, Continuous, Split} and the setting that
chooses it. Only Paged is wired up so far — the other two land next; this
commit is the pref, the state plumbing and the chooser.
One picker rather than two switches ("vertical scrolling" + "month with
agenda"): independent toggles would multiply into four combinations, most of
which nobody asked for. Split does not disable the Agenda view — that stays a
forward multi-day window, while the split pane lists one selected day.
The chooser lives on the Views settings screen (per-view layout belongs with
the other view configuration) and is a hand-rolled FullScreenPicker rather than
OptionPicker, which cannot render previews: the three options differ in shape,
which a word like "Continuous" does not convey. Each card carries a schematic
drawn from theme tokens, and selection reads three ways over — border weight,
container tint and a check — so it never rests on colour alone. The cards are a
selectableGroup with Role.RadioButton for screen readers.
Folded into the ViewCustomization holder, which had spare arity; the outer
settings combine is still at its five-flow limit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Groundwork for the month view style setting (#38, #53); no behaviour change.
- Split layoutCalendarWeek out of layoutMonthWeeks so the continuous style,
which streams weeks and has no enclosing YearMonth to slice by, lays rows
out identically. layoutMonthWeeks (also used by the month widget) keeps its
signature and becomes a loop over it.
- Add MonthUiState.Success.instancesByDay: the grid's events keyed by date and
uncapped, so the split style's day pane can list a date without a second
provider query — the month grid range already covers it.
- Move AgendaDayHeader / AgendaEmptyDayRow / AgendaEventRow and their label
helpers into AgendaRows.kt as internal, so the split pane reuses the agenda's
row vocabulary instead of growing a parallel one. AgendaEmptyDayRow takes its
text as a parameter now that it serves more than "nothing left today".
- Add the month package's first JVM tests, covering week counts, span
continuation across row boundaries, lane stacking and instancesByDay.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review follow-ups to the picker redesign, plus the bugs it exposed.
The live read-out now renders customResult itself — the exact string OK
would save — instead of rebuilding a parallel rule with `interval ?: 1`
and `end ?: Never` fallbacks, which let it confidently describe a rule
that differed from the selected controls whenever a field was invalid.
Shrink the invalid space behind it: a blank amount field reads as its
visible placeholder (1 / 10) rather than as an error, and backing out of
the date picker falls back to "never" instead of stranding a dateless
"on a date". Only an out-of-range 0 remains invalid, and that now says so
rather than greying out OK with no cause.
UNTIL displayed the day after the one picked for zones behind UTC:
toRRule deliberately writes the end of the chosen *local* day expressed
in UTC (the provider applies UNTIL coarsely), so the read side must
convert back before taking the date. Fixes the detail screen too, and
untilLocalDate is extracted so it can be tested.
Also: hoist a remember() out of a conditional (a slot that appears and
disappears breaks positional memoisation), match GroupedSurface's 22dp
corners instead of a drifted local 20dp copy, move the cards onto
GroupedSurface, drop the segmented row's icon slot so longer unit labels
fit, and reserve two lines so the stack stops shifting as the phrase
grows with each weekday.
Extract SettingsPrefs.firstDayOfWeek(scope), replacing three copies that
had drifted onto different initialValues.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Follow-up to 30fcbfa, fixing eight issues found in review:
- Cap the agenda row list at 100. SizeMode.Exact asks Glance for one
RemoteViews per host size where SizeMode.Single produced exactly one,
roughly doubling the payload; with the range reaching
AgendaRange.MAX_CUSTOM_DAYS (365) an uncapped list could push past the
binder transaction limit and the host would show "Problem loading
widget". A trailing day header stranded by the cut is dropped.
- Loosen the height cap so it only catches genuinely squashed widgets.
It previously required ~320dp of height for LARGE, so widening a widget
without also making it unusually tall — the exact resize #51 reports —
stayed REGULAR or COMPACT and the feature was near a no-op for it.
Thresholds now work on height minus header chrome.
- Lock the event stripe to the system font scale. It is a Dp beside sp
text, so at large accessibility settings the text outgrew it and it
under-ran the row it marks.
- Route the day-header and placeholder padding through the metrics table
so vertical rhythm holds at the larger tiers, and derive the text
indent from the row constants instead of a hardcoded 19dp.
- Share the bucketing as widget/WidgetScale.kt so MonthWidget (already
SizeMode.Exact) can adopt one rule rather than growing a parallel copy.
- Anchor the type ramp to Material 3 type-scale roles per CLAUDE.md, with
the two off-scale values marked and justified inline. COMPACT is
unchanged, so a default-sized widget still looks exactly as before.
- Tie the "default size unchanged" test to the provider XML's declared
3-cell band rather than one measured 222dp point.
- Hang the metrics off an ordinal-indexed table so lookup allocates
nothing per recomposition.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Correctness:
- Recurring writes: move the series DTSTART by the *wall-clock* shift applied
to the edited occurrence and re-resolve it in the event's zone, instead of by
a millisecond delta. The old delta baked in whichever UTC offset applied on
the edited occurrence's date, so pinning a recurring event to another zone —
or editing an occurrence on the far side of a DST boundary from the series
anchor — shifted the whole series by an hour. Also snaps the anchor to a UTC
midnight when the event becomes all-day.
- Detail card and edit form now resolve a pinned zone's abbreviation/offset at
the *event's* instant, not at "now", so a July event no longer reads
"CET · 10:00 AM" when opened in January.
- Agenda: the zone used to label multi-day rows now travels on
AgendaUiState.Success rather than a process-lifetime file-level constant, so
labelling can't disagree with the grouping after a device time-zone change.
- Week title: spell out the year when the week straddles New Year, via a new
forceYear flag on formatCalendarTitle.
Performance:
- Build the ~600-entry zone catalogue off the main thread (produceState +
Dispatchers.Default); resolve the device row's summary on its own so it still
renders complete on the first frame.
- Pre-normalize each TimeZoneOption's search keys at construction, turning
~2400 NFD normalizations per keystroke into plain prefix/substring checks.
Hoist the combining-mark Regex out of the hot path.
- Key the edit form's local-time line on the fields it reads instead of
recomputing it on every keystroke.
- Move LauncherNameManager's PackageManager binder calls off the main thread.
Cleanup:
- Drop a duplicate Public icon import and the unused
event_edit_timezone_clear string.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The "Upcoming" agenda widget used Glance's default SizeMode.Single: it was
composed once at the minimum size and the launcher stretched that single
RemoteViews when enlarged, so the text stayed small-widget-sized no matter how
big the widget grew. Reported as a "font size" request (#51), but it's really a
missing size-response.
Switch to SizeMode.Exact (like MonthWidget) and read LocalSize.current to pick
one of four tiers (COMPACT/REGULAR/LARGE/XLARGE), scaling type and row metrics.
Exact over Responsive so the ~30-day LazyColumn isn't replicated per tier.
Width picks the tier, height can only lower it. Width governs how much of a
title fits on a row, so it's what should drive type size; height only decides
how many rows are visible, so a tall narrow widget shows more events rather than
bigger text. Height does act as a cap, though, or a squashed widget would keep
the large type its width earned in a sliver of space. Thresholds are spread over
the width range a phone actually produces (measured on a Pixel/Nova: a compact
widget is 222dp wide, a large one 378dp) rather than a theoretical range, so the
tiers are reachable in practice; XLARGE is reserved for tablets/foldables.
COMPACT reproduces the original constants verbatim, so an existing widget is
visually unchanged. The tier logic lives in a pure, Glance-free AgendaScale.kt
(compose.ui.unit only) and is JVM-tested: the COMPACT baseline, the width
buckets, the height cap stepping a squashed widget down, and that height never
raises the tier.
No new setting: the widget follows the size the launcher/user already chose.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a Settings → Appearance toggle that switches the app's launcher label
between "Calendula" and "Calendar", for users on launchers that can't rename
apps themselves.
The launcher entry moves off MainActivity onto two <activity-alias> components
(DefaultNameAlias / CalendarNameAlias); exactly one is enabled at a time via
PackageManager.setComponentEnabledSetting. MainActivity keeps every other intent
filter; the android.app.shortcuts meta-data moves onto both aliases so the
long-press shortcut still publishes. Component-enabled state is the single source
of truth — no persisted preference.
The ComponentName uses the applicationId for the package (carrying the
.debug/.releasetest suffix) and the namespace for the class, since manifest
".Alias" names resolve against the namespace; switching to Calendar enables the
target alias before disabling the other to avoid a zero-entry launcher transient.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The agenda window starts at the anchor's local midnight, which east of UTC
is the previous day's 22:00 UTC. All-day events live at UTC midnights with an
exclusive next-midnight end, so yesterday's all-day event (a birthday, say)
still overlaps today's window start and is returned by the provider.
groupAgendaDays then clamped its first day up to the anchor and, via a trailing
coerceAtLeast(firstDay) on the last day, pulled it onto the anchor's "today"
section — the multi-day fix only stopped the forward leak on interior days.
Drop instances whose true last day (resolved in UTC for all-day events) falls
before the anchor, or whose first day falls past the window end: they occupy no
visible day and must not be clamped onto an edge. Adds eastern-zone regression
tests for both edges.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Code-review follow-up.
- The multi-day expansion in groupAgendaDays is shared with the agenda
widget, but only the screen's summary was made day-aware — so the
widget rendered the raw "start – end" on every spanned day, the very
bug the screen fix cured. Hoist a pure agendaTimeLabel(event, day,
zone) into the shared agenda layer and resolve strings from it in both
the screen and the widget, so they label identically. (findings 1, 2)
- groupAgendaDays could silently drop an instance whose clamped span was
empty (firstDay > lastDay); floor lastDay at firstDay so a returned
instance always surfaces on at least its first visible day. (finding 3)
- agendaTimeLabel resolves the span days once instead of the summary
recomputing them 2–3× per row. (finding 5)
Finding 4 (within-day sort) needs no change: sorting by absolute start
already places a still-running multi-day event at the top of each day it
continues into, which is chronologically correct (it is ongoing from
that day's midnight), and at its real start time on its first day.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Review follow-up. spanFirstDay/spanLastDay resolved every event in the
device zone, but all-day events live at UTC midnights with an exclusive
end — east of UTC (e.g. Europe/Berlin) that pushed spanLastDay onto the
next day, so a single-day all-day event reported spansMultipleDays and
leaked onto a second agenda day. Resolve all-day dates in UTC, matching
the Week view and detail card. Adds eastern-zone regression tests that
the prior UTC-only tests could not catch.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Search matched city, id, and the long localized name but not the
abbreviation, so typing "CEST" found nothing. Match it too: an exact
abbreviation hit ranks just under a city prefix, so typing an abbreviation
gathers every zone that shows it (all the CEST zones at once). It matches
the region-resolved abbreviation — i.e. exactly what the row displays.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
groupAgendaDays keyed each instance by its start day alone, so a
multi-day event surfaced only on its first day and vanished from the
rest of its span in both the Agenda screen and the agenda widget.
Expand each instance across every day from its start (clamped to the
anchor for ongoing events) through its last occupied day, bounded by the
visible window end. An event ending exactly at midnight — including the
exclusive next-midnight all-day events end at — does not reach that
boundary day.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"Central European Time (Europe/Berlin)" is too wide — it made the zone
field wrap and stretch. Lead with the id ("Europe/Berlin") and follow it
with the abbreviation instead: "CEST · GMT+02:00" in the picker and edit
card, "CET · 8:00 AM – 9:00 AM" on the detail card (abbreviation + the
event's own-zone time).
The abbreviation is resolved DST-aware at the same instant as the offset
(CET vs CEST) via "zzz"; zones with no named abbreviation fall back to a
"GMT+05:30" form, and zoneDescriptor drops the separate offset in that
case so it isn't stated twice. The long localized name is kept on the
option for search only — typing "pacific" still works — but no longer
shown.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two review fixes.
The zone picker's search box was a raw Material OutlinedTextField — the
only one left in the app, and against the convention DialogControls
states outright ("the family's InlineTextField over a tonal surface, not
Material's outlined field"). Rebuild it on InlineTextField over a tonal
surface, with the clear button inside the surface since the picker's top
bar is the title rather than a search field.
Showing a pinned event only in its own zone answered "what was it set
to?" while dropping "when is it for me?" — the user had to do the offset
arithmetic. Show both whenever they differ:
- the edit form keeps editing the event in its own zone (that's the time
it was set at) and captions it with the local equivalent;
- the detail screen keeps local times primary and now leads the zone card
with the original ("8:00 AM – 9:00 AM in New York") instead of naming
the zone and nothing else.
EventForm.timesIn is pure, so the conversion — including crossing the
date line and each zone's own DST, which don't move together — is a
plain JUnit test rather than something only reviewable on a phone.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Every write sampled ZoneId.systemDefault() and stamped it into
EVENT_TIMEZONE, so the column was real but only ever held the device's
zone: an event synced from elsewhere could be read in its zone, never
authored in one.
Give EventForm a nullable `timezone`, where null keeps meaning "the
device zone at save time" — so every existing call site behaves exactly
as before — and a non-null value pins the event to a zone it then tracks
across DST. toWriteTimes resolves the form's zone ahead of the device's;
toEditForm pins only when the stored zone differs from the device's, and
prefills such an event in its own zone so the form shows the wall-clock
the event actually means.
Two provider-contract bugs fall out of this:
- Editing the time of a foreign-zone event rewrote EVENT_TIMEZONE to the
device's. The instants stayed right, so nothing looked wrong, but the
event silently stopped tracking its zone and would drift an hour at the
next DST boundary. Only the timesChanged gate spared title-only edits.
- A zone change with an untouched wall-clock is still a time change (the
same 09:00 elsewhere is a different instant), so it now trips
timesChanged and rewrites DTSTART instead of being dropped.
All-day events keep carrying no zone at all: they're date-anchored, and
the UTC midnights they normalise to are an anchor rather than a location.
TimeZoneCatalog is pure JVM so the search ranking and DST-aware offsets
stay plain JUnit tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add an "Always show today" setting (Settings → Agenda, on by default) that
keeps today as the first entry in both the Agenda screen and its home-screen
widget even once nothing is left today. Under today's normal header a small
"No more events today" card appears — the coffee-cup empty-state motif in the
app, a rounded surface in the widget — so the first rows you see are clearly
today's rather than a future day's.
The anchor is a pure, JVM-tested helper (anchorTodayIfMissing) applied after
past-event filtering; in-app it only kicks in when the window starts on today,
never on a jumped-to date. The widget reads the pref reactively via per-instance
Glance state, mirroring the range/past-event settings.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A reminder fired ahead of an event on a different day showed only the
event's time (e.g. "09:30 – 10:00"), making it look like it was
happening today. reminderTimeText now prefixes timed events with a
relative day: "Tomorrow"/"Yesterday", the short weekday for another day
this week, or the exact date for anything further out.
The this-week boundary honours the user's "week starts on" setting: the
resolved first day of week is threaded through from ReminderNotifier, so
e.g. a Sunday reads as next week under a Sunday-start locale. All-day
events keep their explicit date (never ambiguous), and cross-midnight
timed events keep both explicit dates.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The calendar row in the editor is now tappable when editing an existing
event: picking a different calendar moves the event there on save, rather
than forcing a delete-and-recreate.
CALENDAR_ID is sync-adapter-owned and can't be updated in place, so the
move is copy+delete: the master row is re-inserted on the target calendar
(preserving UID_2445 so backup dedup and sync identity survive), its
reminders and editable guests are copied, and — for a recurring series —
every exception is replayed against the new master (modified occurrences
via CONTENT_EXCEPTION_URI, cancellations as STATUS_CANCELED). The user's
field edits are then applied with the normal series update. Everything on
the new side is built before the source is deleted (post-before-delete),
with a rollback of the copy on any failure, so a move is all-or-nothing.
A calendar change forces whole-series scope, so it skips the recurring
scope dialog. Managed special-dates calendars stay locked. Colour is not
carried across (a raw/keyed colour may be invalid on the target account),
matching the existing calendar-switch behaviour.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The cancelled-exception fix works on synced calendars but not local ones. A
cancelled exception only attaches to its parent through ORIGINAL_SYNC_ID; a
local event has no _sync_id, so the link never forms and the provider's
expansion of the *parent* collapses — every other occurrence disappears, which
is the original #47 corruption, just on a different calendar type. Verified
on-device both ways: a DAVx5 series survives a single-occurrence delete, the
same series on a LOCAL calendar vanishes entirely.
deleteOccurrence now branches on _sync_id. Synced events keep the (verified)
exception path. Events without one — local calendars, and synced events not yet
pushed — add the occurrence to the master's EXDATE, which needs no parent link
and is the canonical iCalendar way to drop one; a sync adapter carries it
upstream unchanged if the calendar later syncs.
Two provider quirks shape the write (both observed on a Pixel):
- An EXDATE-only update is not treated as a recurrence change: the expanded
Instances rows are left alone, so the occurrence stays visible. The
time/recurrence set has to ride along to force re-expansion.
- DTSTART alone is worse — the provider then recomputes lastDate as if the event
were a single instance and collapses the series to its first occurrence.
DTSTART + DURATION + RRULE + zone together re-expand it correctly.
This path is reached in normal use: Calendula's own contact special-date
calendars are local and hold all-day yearly series, so deleting one birthday
occurrence went through it. All-day series take the VALUE=DATE EXDATE form.
Adds pure buildOccurrenceExdateValues + JVM tests (timed, append, duplicate
fold, all-day).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A recurring series row carries DURATION, not DTEND, so EventDetailMapper's
end == begin fallback rendered it zero-length. That was harmless while every
caller supplied per-occurrence times from Instances, but the bare
content://…/events/<id> VIEW intent added in #48 names no occurrence and
keeps the row's own times — so a series opened from a third-party widget
without begin/end extras showed as "10:00 – 10:00".
Read DURATION in the detail projection and derive the end from it, the same
way SearchMapper and IcsExportMapper already do.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CalDAV sync adapters (DAVx5) publish all ~147 CSS3 named colours into
CalendarContract.Colors, so the event-colour picker showed a full screen
of alphabetically-scrambled, partly duplicated swatches.
Curation now runs in the space the picker actually paints — every swatch
is softened through pastelize, which pins lightness and caps saturation,
so the raw palette's lightness axis is invisible on screen. Judging
distinctness there: colours that paint identically collapse to one
(folding aliases, dark/light shades of a hue, and the neutrals together),
oversized palettes drop washed-out neutral-origin tints and thin by CIE76
ΔE in painted Lab, and survivors sort continuously by painted hue with the
wheel cut at its single widest gap. The CSS3 dump lands at ~33 distinct,
rainbow-ordered swatches; small hand-picked palettes (Google's) pass
through untouched. Every surviving swatch keeps its provider colour key so
picks still round-trip through sync.
This revives work stranded on fix/caldav-color-picker (never merged) and
adapts it to the floret-kit extraction of pastelize: the curation's
painted-space transform now lives self-contained in domain/pastelArgb as a
mirror of floret's pastelize shaping, rather than the two sharing one
function.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"Delete only this event" on a recurring series wrote a cancelled
exception carrying just ORIGINAL_INSTANCE_TIME + STATUS_CANCELED. Without
DTSTART + DURATION the provider clones the master *with its RRULE intact*
and cancels the whole clone, so every other occurrence vanished, the
target survived as a "cancelled" ghost, and re-deleting toggled the
series back — exactly the reported corruption.
Anchor the exception as a single instance (DTSTART + DURATION + zone +
all-day, read from the series row) so the provider clears the inherited
RRULE and cancels only that occurrence — the same discipline the edit
path already documents (Codeberg #16). Also filter STATUS_CANCELED out of
the instances grid query so the cancelled occurrence disappears instead
of lingering as a tappable ghost (NULL status is kept — a normal event).
Extracts the exception ContentValues into a pure buildOccurrenceCancelValues
helper with JVM tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Brings the 2.14.0 feature set (Month week numbers, .ics restore +
per-calendar export picker, Day view on date-header tap, wrapped long
titles, Simplified Chinese) onto the floret-kit migration without
resurrecting the inline component copies the kit now owns.
Conflict resolutions (both were import-block collisions where the
migration repointed to de.jeanlucmakiola.floret.* while release kept the
old app-local imports):
- CalendarsScreen.kt: keep the kit imports (CollapsingScaffold,
DialogAmountField/DialogUnitDropdown, collapseExit/expandEnter,
predictiveBack) and repoint FullScreenPicker + positionOf to the kit
(both moved out of ui.common). Keep the genuinely app-local
LeadingAvatar/SourceLogo/curatedSourcePackage; drop the renamed
calendarCollapseExit/calendarExpandEnter (0 uses).
- ImportScreen.kt: keep the app-local CalendarPickerGroups, use the kit's
predictiveBack, drop the now-unused OptionCard import.
Semantic fixup: CalendarPickerGroups.kt (new in 2.14.0) relied on
same-package resolution of GroupedRow/Position, which the migration moved
to the kit — added the explicit floret.components imports.
Verified: :app:compileDebugKotlin, ./gradlew test (app + kit), and
scripts/check_translations.py all green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Supersedes the stale chore/floret-kit-core-time branch: re-applies the
floret-kit migration on top of current main (122 commits ahead of the old
branch's base), pinning the kit at the multi-value-reminders + pinned-picker HEAD.
- Submodule + composite build (includeBuild), 6 module deps, CI submodules:
recursive, reproducible-release scan extended to the kit, F-Droid recipe.
- Deletes the inline copies now owned by the kit (GroupedList, Picker scaffolds,
InlineTextField, OptionCard, DialogControls, CrashReporter + dialog/submit,
OnboardingScaffold, AppLanguage, TimeBridge, ReorderableColumn, DebugRibbon)
and redraws them from components/identity/core-crash/core-locale/core-time.
- Reminder overrides drawn from core-reminders (multi-value ReminderOverride +
codec); Calendula keeps its app-specific bits (all-day resolution, labels,
presets, the multi-select ReminderDefaultPicker, its own CrashReportActivity).
- Theme draws FloretExpressiveTheme while keeping the user-typography param.
Build pending (deferred): run ./gradlew :app:compileDebugKotlin with ANDROID_HOME
(or floret-kit/local.properties) set.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add an opt-in left gutter to the Month grid showing the ISO calendar-week
number, gated by a new "Week numbers" display setting (default off). The
number is computed on each row's first day — the same basis as the Week
view's badge — so the two views agree, and rendered as a low-emphasis
onSurfaceVariant label so it recedes across all six rows rather than
competing with the event bars. The weekday header reserves a matching
gutter so the day columns stay aligned.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The fake now records the calendarIds it receives; two repository tests
assert exportEvents forwards a chosen subset and defaults to null (all
eligible calendars), closing the coverage gap for the per-calendar
export selector.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Tapping Export with more than one exportable calendar now opens a picker
to choose which local calendars to include (all selected by default); a
single calendar exports straight through as before. Threads an optional
calendarIds filter through exportEvents/exportableEvents (null = all
eligible), so the auto-backup path is unaffected. The backup section is
now gated on there being at least one exportable (non-managed) calendar.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
toEventDetailCore returned null when a present DTEND preceded DTSTART, the
only remaining false-drop that surfaces as the generic "Something went
wrong." error screen — the same un-openable trap as the pre-1970 DTSTART
bug, and worse because the user can't even open the malformed event to fix
it. Clamp the end to DTSTART (a zero-length event) instead, matching how
SearchMapper already coerces its end. After this the detail mapper drops a
row only when DTSTART is genuinely absent (unrenderable).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Register an intent-filter for ACTION_INSERT on the events dir mime type
(vnd.android.cursor.dir/event), the way the AOSP calendar accepts one, so
other apps and widgets (e.g. the Todo Agenda widget) can launch Calendula
to create a new event.
MainActivity.insertFormOrNull parses the standard CalendarContract extras
(EXTRA_EVENT_BEGIN_TIME/END_TIME/ALL_DAY, Events.TITLE/DESCRIPTION/
EVENT_LOCATION/RRULE) into a prefilled EventForm via the pure, unit-tested
buildInsertEventForm — omitted fields fall back to the same defaults the
in-app "new event" uses (next full hour, +1h). The form is routed through
the existing single-event prefill channel (RootScreen → CalendarHost →
the create form for review), with calendarId left null so it resolves to
the last-used / first-writable calendar. No new permission is needed
(WRITE_CALENDAR is already held), and the user still explicitly saves.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
